The ifcfg-rh implementation should be similar to the one from keyfile.
Rename the variables and function that have the same meaning.
Do this trivial commit first, before starting refactoring.
When adding a connection to NMSettings we did not check for
duplicate connection UUIDs (which could for example happen
if two different plugins report a conflicting UUID).
Also, we would not check that an already added connection
changes it's UUID.
Both could lead to have duplicate connections (by UUID).
Avoid that two ways:
- when adding a connection to NMSettings, ensure that we don't add
a conflicting UUID. Otherwise just bail out and do nothing.
- when modifying a connection that is already added to NMSettings,
enforce that the UUID cannot change. Otherwise fail with error.
For ifcfg-rh plugin this situation still can happen during reload.
In this case error out and refuse to update the connection. After
all, the user configured invalid UUIDs.
https://bugzilla.redhat.com/show_bug.cgi?id=1171751
Also during add_connection() we must take special care of not "adding" a
connection with a conflicting UUID. In that case we want to fallback to
"update".
update_connection() already does all the checks, so call
update_connection() from add_connection().
If there are keyfiles with duplicate UUIDs, read_connections()
would iterate over the files, loading them as they appear and
overwriting duplicate connections that were just loaded.
For example, have keyfiles 'A' and 'B' with the same UUID.
On start, NM might first load 'A', then 'B'. 'B' would replace the
content of 'A' which was just loaded.
On reload, NM would first overwrite 'B' with 'A', and then again
overwriting 'A' with 'B'.
Fix that by accept the first found connection and don't overwrite
it during the same read_connections() run.
Also sort the files by file modification timestamp so that we
get a reproducible and sensible behavior.
new_connection() and update_connection() are very similar as both
must anticipate collisions of UUIDs.
When reloading a connection (update_connection(), previously), the loaded
connection for a certain path might actually replace another existing
connection. In this case, the old connection must be removed, and
the existing one updated instead.
If reloading a connection changes the UUID to a new value, the old
connection must be removed likewise and a new connection added.
Merge both functions into update_connection().
Only log connection diffs when we update a connection that we actually
care about.
Note that most plugin specific connections use
nm_settings_connection_replace_settings() in their constructor
to initialize themselves. These occurrences are not interesting
and spam the logfile.
conn_name can be NULL now as the connections are tracked by UUID
(see commit 689dadaffb)
Sep 29 21:20:34 Jdeapad NetworkManager[1339]: <info> getting unmanaged specs...
Sep 29 21:20:34 Jdeapad NetworkManager[1339]: <info> Checking unmanaged: (null)
Sep 29 21:20:38 Jdeapad kernel: NetworkManager[1339]: segfault at 0 ip
00007f1bfffbedd0 sp 00007fff98daa628 error 4 in
libglib-2.0.so.0.4002.0[7f1bfff86000+130000]
https://bugzilla.gnome.org/show_bug.cgi?id=737645
refresh_object() raised a spurious change event for the route we
are about to delete. Suppress that by adding an internal reason flag.
Fixes: 41e6c4fac1
On resume configured interfaces are unmanaged to clear their pre-resume
state and then re-managed. Eventually the interface should end up moving
to the DISCONNECTED state, which should trigger an auto-activate check in
the Policy. If connectivity checking was enabled, that auto-activate check
would fail because the Manager's state was still NM_STATE_ASLEEP.
This caused bridge slaves not to auto-activate on resume, which left bridges
without connectivity.
The manager never left NM_STATE_ASLEEP when connectivity checking was
enabled due to nm_manager_update_state() returning early when kicking
off a connectivity check. Instead, the manager's state should always
be updated to accurately reflect the current state.
https://bugzilla.redhat.com/show_bug.cgi?id=1162636https://bugzilla.gnome.org/show_bug.cgi?id=742675
If a connection has an associated "rule-NAME" or "rule6-NAME" file,
don't try to read in the routes, since NetworkManager won't be able to
parse them correctly. Instead, log a warning that they will need to be
applied via a dispatcher script, and provide a script that would do
that in examples/dispatcher/.
Add a "filename" property to NMSettingsConnection, and set it from
keyfile and ifcfg-rh (replacing the existing priv->path variables in
those connection types). (The other plugins either don't use files, or
don't use per-connection files.)
Don't have the singleton instance of NMDBusManager owned by
the main function. Instead use NM_DEFINE_SINGLETON_DESTRUCTOR()
which also logs what's happening.
Also move the initilization of the instance into the constructed()
method.
NMAgentManager now owns a reference to the DBUS manager and Auth
manager and the dispose() function properly unregisters itself from
both.
Refactor NMInotifyHelper to implement the singleton getter using
NM_DEFINE_SINGLETON_GETTER().
For one this means that the getter no longer increments the reference
count. This was anyway wrong, because no caller of nm_inotify_helper_get()
unrefered the returned reference, hence leaking the singleton.
Also, the getter can no longer fail to create the singleton instance.
Note that none of the callers actually coped with a failure to get
the singleton.
Instead return an instance that does nothing.
One downside (upside?) of this is that we only try once to initialize
the inotify handle.
Also no longer increment the reference count in the getter and
properly disconnect the signals in NMManager:dispose().
Also use the defines for the signal names instead of plain strings.
Deleting routes with metric 0 might end up deleting other
routes with a different metric.
Workaround this in platform to only delete a route with
metric 0 if such a route can be found prior to deletion.
Don't only look into the cache (which might be out of date).
Instead refetch the route we are about to delete to be sure.
There is still a race that we might end up deleting the wrong
route.
https://bugzilla.gnome.org/show_bug.cgi?id=741871https://bugzilla.redhat.com/show_bug.cgi?id=1172780
This match-any behavior ignoring metric is nowhere used. And even if we
would need such a behavior, using 0 is wrong because IPv4 routes can
have a metric of zero.
Handling a route with metric 0 effectively means
a metric of 1024 (user default). Adjust the add(),
delete() and exist() functions to consider routes
with metric 0 as 1024.
This is really a no-op because the directory will be created by
nm_main_utils_check_pidfile() anyway, but make sure the permissions
are synchronized. Now that resolv.conf is symlink into NMRUNDIR,
if normal users don't have permissions to read NMRUNDIR then
DNS won't work.
